Can Readers Tell When Content Is AI-Written? Usually, Yes
People rarely accuse a post of being written by AI. They just scroll past it slightly faster, and read the next one from that account with a bit less attention. That is what ai written content detection looks like in the real world, and it happens well before anyone consciously decides anything.
Readers are not detecting facts. They are detecting shapes, the same handful of sentence patterns turning up post after post, and once the pattern registers the writing stops feeling like a person talking.
This article was drafted with AI help and then edited hard, which is the only honest way to write it. What follows are the specific tells, with the before and after in each case.
The long dash where a comma would have done#
The single most reliable giveaway. AI drops the em dash into sentences at a rate no human writer sustains, usually to bolt an afterthought onto a finished clause.
Before: We replaced the board that afternoon - the owner had been waiting a month.
After: We replaced the board that afternoon. The owner had been waiting a month.
Two short sentences almost always beat one sentence with a dash in the middle. Search your draft for the long dash character and remove every one of them. Nothing is ever lost.
Three of everything#
Sets of three are the rhythm the model falls into when it does not have three distinct things to say. The third item is usually a restatement of the second in different words.
Before: Our team delivers fast, reliable and professional service across Auckland.
After: We turn up when we said we would, and we do not leave a mess.
The tell is not the number itself. It is that the three items carry roughly one item's worth of information. Cut to two and make both concrete, and the sentence starts doing work.
The reversal formula#
"It's not about X, it's about Y." "This isn't a tool, it's a partner." The construction sounds like a conclusion being reached, which is why it is seductive, and it delivers nothing that the plain statement would not.
Before: It's not just a website, it's your best salesperson.
After: A good site answers the two questions people ring you about, so fewer of them need to ring.
Same idea, no theatre, and the second version tells the reader something they can act on.
Bullets that lead with a bolded word and a colon#
Instantly recognisable, and increasingly a signal that nobody edited the output. The bolded word usually repeats a category label rather than saying anything.
Before: Speed: Our AI responds in seconds.
After: Enquiries get an answer in under a minute, including at nine on a Sunday night.
Write bullets as sentences. If a bullet cannot survive without a label in front of it, the label is carrying the meaning and the bullet is not.
Praise with nothing underneath it#
Exceptional, seamless, unparalleled, cutting-edge, robust. These words are how the model fills space when it has no specific detail available, because a superlative is always safe.
Before: We provide exceptional customer service and unparalleled attention to detail.
After: We ring you back the same day, and if we are running late you get a text before we are late, not after.
The test is whether a competitor could publish the sentence unchanged. If they could, delete it and replace it with something that only applies to you.
Throat-clearing before the point#
The opening line that announces a topic rather than starting it. "In today's digital landscape." "It's worth noting that." "Let's explore why this matters."
Before: In today's competitive market, having a strong online presence is more important than ever for small businesses.
After: Most people who ring you have already read your website and decided you look expensive.
Delete the first sentence of any AI draft and read what is left. The piece is nearly always better, which tells you what the sentence was doing.
The rhetorical question it answers immediately#
"So what does this mean for your business? It means faster responses." The question adds a beat of drama and no information, and three of them in one piece reads like a keynote nobody asked for.
Before: Why does response time matter? Because customers expect answers now.
After: Someone waiting two days for a quote has already booked whoever answered first.
Where this actually costs you money#
Nobody cancels a booking because they spotted a formulaic sentence. The damage is slower and it comes in two forms.
The first is attention. A feed of average-sounding posts trains your audience that your account is skippable, and once that is learned it is expensive to unlearn. The second is credibility on price. A business that writes in unbacked superlatives has given a shopper no reason to pay more than the cheapest quote, because it has not demonstrated anything specific.
The fix in both cases is the same, and it is one detail per post. What the job was, what went wrong, what it cost, what you would tell a friend to check. A model cannot invent that, which is why it reaches for adjectives instead.
There is a third cost that shows up later, in the inbox. When the writing on your website and your posts is generic, people arrive at the enquiry stage without a clear idea of what you do or what you charge, so you spend more phone time filtering. Specific writing sorts people before they contact you, and the businesses that do it well tend to field fewer enquiries and win a higher share of them.
The SEO question, answered plainly#
Google has been consistent that it evaluates content on whether it is helpful and original, not on the process that produced it. Content built mainly to manipulate rankings gets demoted, and that has been true since long before AI writing tools were widely available.
The practical version for a local service business: a thin page that restates what every competitor already says will struggle whether a human or a model wrote it. A page that answers a real question with real specifics will hold up either way. The ranking factors worth your attention are covered in SEO for service businesses.
AI written content detection tools are not reliable enough to act on#
Owners ask about detector software constantly, usually because a client or a marketing person has waved a score at them. Those tools return false results in both directions often enough that no decision should rest on one, and plain, formulaic human writing gets flagged regularly.
Treat a detector score as a hint that the writing is generic, which is a real problem worth fixing, rather than as evidence about how it was produced. Your own ear, reading a draft aloud, is the better instrument and it is free.
The editing pass, in five minutes#
Take any AI draft and run this before it goes anywhere.
- Delete the first sentence. Read what is left and see whether you miss it.
- Remove every long dash. Replace with a full stop.
- Find every set of three and cut it to two, making both specific.
- Delete every superlative that is not attached to evidence.
- Add one detail from a real job that nobody else could write.
- Read it aloud. Anything you would not say out loud gets cut.
Step five is the one that matters and the one people skip. The other four remove the tells. Only that one adds a reason to keep reading.
Doing this well is much easier once you have written down how you actually sound, which is a one-hour job covered in teaching AI to write in your voice. Where AI belongs in the wider content workflow, including the stages it handles better than the drafting, is set out in AI social media management. If your problem is having nothing specific to write about in the first place, that is a sourcing problem rather than a writing one, and the process for fixing it is in building a month of content in an afternoon.
Common questions
Can Google tell if content is written by AI?
Google has said it rewards helpful, original content regardless of how it was produced, and demotes content made mainly to game rankings. In practice thin, repetitive pages lose visibility whether a person or a model wrote them, and useful pages hold it either way.
Do AI content detectors actually work?
Detection tools produce both false positives and false negatives often enough that no serious decision should rest on their score. Plain, formulaic human writing frequently registers as AI-generated. Human readers judging by feel are more reliable than the software claiming to measure it.
Is it bad for my business to use AI to write posts?
Using AI to draft is fine. Publishing unedited output is the problem, because the default style is generic enough that readers register it as filler. Add a specific detail only your business knows and cut the formulaic phrasing, and nobody minds how the first draft appeared.